A Fault Detection Method based on the Deep Extended PCA – SVM in Industrial Processes

With the stable and safe requirement in the industrial technology, the fault detection, has attracted more and more attention from both scholars and companies. To make full use of the linear and non-linear features for fault detection, a method named deep extended principal component analysis - support vector machine (Deep EPCA-SVM) was proposed by combing the PCA and kernel PCA with the deep structure. Both the PCA and kernel PCA were iteratively implemented in the principle subspace and residual subspace for the extraction of linear and non-linear features. The offline model was built and applied to monitor the fault of Tennessee-Eastman process. The validation performances indicated that our proposed model outperformed the traditional PCA and PCA-SVM model, and showed higher fault detection rate than the deep DPCA-SVM.